Selecting Thresholds for Sequential Fault Detection Tests

Abstract

Many redundancy management (RM) algorithms use sequential tests for detecting and identifying component failures because the sequential processing of noisy data samples often results in significant decreases in the probabilities of false detection and of missed detection relative to single sample tests. However, this improvement in the error probabilities comes at the cost of a larger delay in detecting or identifying a failure. The error probabilities and delay statistics characterize the performance of the sequential tests, and these performance properties play a crucial role in determining the overall fault tolerant system performance. In this paper, a simple technique to derive approximately optimum thresholds for sequential failure detection tests is indicated where the performance of the fault tolerant control system is the optimality criterion. The threshold selection method is illustrated by a generic quadraplex redundant system example.